From sleep staging to spindle detection: a case study on end-to-end automated sleep analysis

Area of Science:

• Neuroscience
• Computational Biology
• Sleep Medicine

Background:

• Automating sleep analysis, including sleep stages and microstructures like sleep spindles, is crucial for large-scale studies.
• Previous research focused on individual analysis steps, leaving multi-step automation feasibility unclear.

Purpose of the Study:

• To evaluate if fully automated sleep analysis using validated machine learning models can replicate expert-based findings in bipolar disorder.
• To assess the feasibility of combining automated sleep staging and spindle detection for complex sleep research.

Main Methods:

• Utilized RobustSleepNet for automated sleep staging and SUMOv2 for automated spindle detection.
• Compared automated analysis results with a previous expert-based study on bipolar disorder patients and healthy controls.

Main Results:

• The automated analysis qualitatively reproduced key findings, showing significant differences in fast spindle densities between bipolar patients and controls.
• Automated methods completed analysis in minutes, a task previously taking months.
• Individual models achieved performance at or above inter-rater agreement for sleep staging and spindle detection.

Conclusions:

• Fully automated sleep analysis shows potential for large-scale sleep research, offering efficiency and consistency.
• Despite quantitative differences, automated approaches can facilitate research by replicating qualitative findings and improving speed.
• Publicly shared code and the SomnoBot platform aim to support future large-scale sleep studies.